Kai retweeted
I attached 20 tokens to "I don't have enough information to answer that." and an LLM judge marked it correct 107 times out of 108 (seed 1)
108/108 (seed 2 and 3)
The judge had the right answer in front of it the whole time 🧵
Kai retweeted
updated the portfolio, have a look again
(lemme know what you think)
vinspace.me
btw i have a "normal" portfolio also:
vinspace.me
Kai retweeted
Philosophical article by @devvaibhav37
Check this out people
Its really good.
medium.com/@devvaibhav37/not…
Kai retweeted
I built a small AI monitor to catch chatbot mistakes.
On mistakes I planted, it caught 89%.
On mistakes real chatbots made, it caught 42%.
Same model, same prompt. What changed was the mistakes 🧵















